Integrated landslide susceptibility and risk mapping using geophysical data in Çayeli (Rize) district
Abstract
In this study, an integrated machine learning-based analysis was conducted to determine landslide susceptibility in Çayeli district. Topographic, hydrological, geological, climatic, and seismic/geotechnical factors were jointly evaluated using data obtained from AFAD, MTA, CORINE, OpenStreetMap, and Digital Elevation Models. In addition, 55 geological-geotechnical reports obtained from the Çayeli Municipality were examined, and geophysical data were integrated into the GIS-based model. A total of 28 conditioning factors were initially generated, and after Variance Inflation Factor (VIF) analysis, variables with high multicollinearity were removed, resulting in 22 final factors. Ten different machine learning algorithms, including Logistic Regression, Random Forest, XGBoost, LightGBM, SVM, and Artificial Neural Networks, were comparatively evaluated. The models were trained using 5-fold cross-validation and validated with an independent test dataset. In the final stage, SHAP analysis was applied to interpret the contribution of conditioning factors to model predictions. Landslide susceptibility maps were produced and showed high spatial agreement with existing landslide occurrences in the study area. The study demonstrated that machine learning and explainable artificial intelligence approaches provide effective results for landslide susceptibility assessment.